The profound influence of microorganisms on human life and global biogeochemical cycles underlines the value of studying the biogeography of microorganisms. Microbial abundance matrices presumably reflect the ecological intricacies of microbial communities but in doing so are prohibitively large, typically including several thousand operational taxonomic units (OTUs)—significantly larger than the number of samples. This makes meaningful statistical and ecological analyses daunting, especially when not all OTUs play an ecologically relevant role in the community—potentially due to low abundance and/or minimalist role in synecological processes. Removal of uninformative OTUs (downweighting) is desirable before proceeding with downstream analyses, though the ideal downweighting approach will depend on the research question and may be confounded by the idea that species rarity does not necessarily equate to ecological irrelevance. Here we propose a new methodology for statistical downweighting based on multivariate metrics that are largely independent of abundance (here referred to as “winnowing”). This research was designed to address the question: can we use statistical properties of microbial communities to isolate the most ecologically influential OTUs (i.e., defined via variance contribution or centrality metrics) and optimize interpretation of treatment effects?
Results/Conclusions
To do this, we used a combination of variance- and centrality-based approaches. We analyzed three 16S rRNA gene sequencing datasets related to the impacts of: plant invasion, soil pH, and plant genotypes. First, the abundance matrices were converted to Hellinger distance metrics to reduce the influence of rare species in the multivariate analyses. Next, centrality and inertia metrics were calculated on the full dataset, then again iteratively after removing N-species with replacement. Permutational Analysis of Variance (PERMANOVA) was used to identify the removal step at which the treatment effect was maximized. We winnowed the three datasets to 1–3% of the original number of OTUs and found this was useful in determining which species were central to differentiating bacterial communities based on a priori identified treatment effects. Though there is no “one size fits all” solution to microbial downweighting, we found by winnowing communities based on multivariate metrics, we were able to discern key players involved in ecological functioning under plant invasion, different soil environments, and plant genotypes. Techniques like winnowing may prove invaluable in future studies of microbial communities, enhancing objective interpretation of DNA sequencing datasets.